OpenAI CFO unveils 4-metric scorecard for measuring AI ROI

Sarah Friar, CFO of OpenAI, has articulated a new framework for measuring whether artificial intelligence investments deliver genuine economic value. Publishing her scorecard in a blog post, Friar argues that the basic economic question for CFOs and business leaders is whether the value of work AI completes grows faster than the cost of producing it. This represents a departure from how software success has traditionally been measured: through adoption metrics such as the number of active users, seat count, and renewal rates. AI, Friar contends, must be measured by the actual work it accomplishes.

At the core of Friar's framework is a metric she calls "useful intelligence per dollar," which has four elements. First: Is the AI completing work that matters? Second: What does each successful task cost? Third: Can people depend on the result? And fourth: Does each dollar produce more value as usage grows? In practice, she explains, leaders should track the volume of AI-completed work that meets a defined quality bar, total the full cost of that work, and divide by the number of successful tasks to arrive at a cost per successful task. The test is then whether people can reliably depend on the output and whether, over time, high-quality completed work grows faster than total cost while quality holds or improves. If this occurs, each AI dollar produces more value. Compute sits at the center of this equation, Friar notes, and OpenAI's job is to make it better with every generation: "more capable models, faster and more dependable results, and lower costs for the work customers need done."

For OpenAI, a large cloud provider (hyperscaler), compute is not simply a technology expense but a strategic asset. The company does not publish formal capital expenditure guidance as a private firm, but in January 2025 it announced the Stargate initiative, outlining a plan to invest up to $500 billion（約80兆円） over roughly four years to build large-scale AI infrastructure in the United States. The initial phase targets about $100 billion（約16兆円）, with the broader buildout accelerating toward a 10-gigawatt capacity goal in the U.S. by 2029. According to reports, OpenAI's initial public offering could occur as soon as this summer or as late as 2027, with the company already valued at $852 billion（約140兆円） and approaching the $1 trillion（約160兆円） range.

Friar's framework reflects a broader organizational shift. While finance chiefs have historically led capital allocation and investor communication, they are increasingly expected to help determine strategy, including placement of long-term bets such as AI spending, alongside the CEO. McKinsey's 24th annual Global CFO Forum, which brought together about 100 finance chiefs from over 30 countries representing some of the world's largest organizations, captured this transformation. Andy West, a senior partner at McKinsey and global co-leader of the firm's Strategy and Corporate Finance practice, conducted an informal poll asking CFOs whether the strategy function reports to them. About two-thirds raised their hands—a sharp increase from five years ago, when less than a third would have answered yes. At the forum, last year's conversation centered on CFOs experimenting with AI; this year, discussion shifted decisively toward enterprise-wide transformation, West noted.

OpenAI CFO Sarah Friar's scorecard represents a fundamental shift in how enterprise leaders are expected to measure the success of artificial intelligence investments. Historically, software has been evaluated through adoption metrics—active users, seat count, renewal rates—metrics that track engagement rather than economic impact. Friar's framework breaks from that tradition by grounding AI evaluation in actual work output and cost-per-outcome, reflecting the economics of a technology that must justify its computational and financial expense on a task-by-task basis.

This shift is part of a broader reorganization of corporate power structures. At McKinsey's recent Global CFO Forum, a senior partner reported that roughly two-thirds of CFOs surveyed now have the strategy function reporting to them—a significant jump from less than one-third five years ago. This centralization of strategic decision-making in the finance office signals that AI investment decisions, once purely technical, have become capital-allocation questions. CFOs are now expected to weigh long-term AI bets—compute infrastructure, model development, talent—with the same rigor they apply to plant and equipment.

OpenAI itself exemplifies this shift. As a private company, it does not publish formal capital guidance, but the Stargate initiative announced in January 2025 reveals the scale of its compute bet: up to $500 billion（約80兆円） over roughly four years for U.S. infrastructure, with an initial phase of about $100 billion（約16兆円） and a target of 10-gigawatt capacity by 2029. The company's trajectory—already valued at $852 billion（約140兆円） and approaching the $1 trillion（約160兆円） range, with an IPO possibly coming as soon as this summer or as late as 2027—demonstrates that the ability to rationalize compute expense through productivity gains is now central to corporate valuation. Friar's scorecard, in this context, is not merely a measurement tool; it is a language for justifying the capital intensity that AI infrastructure demands.